A Paralyzed Man Just Typed 22 Words Per Minute With His Brain. That's 81% of Your Smartphone Speed.

Twenty-two words per minute. That is how fast participant T18, a man with tetraplegia from a spinal cord injury, typed using only his brain. No finger movement. No eye tracking. No head switch. He imagined tapping keys on a QWERTY keyboard, and a neural decoder translated his motor cortex signals into text on a screen. His speed, reported in Nature Neuroscience on March 17, 2026, is 81% of the average smartphone typing speed for able-bodied users in his age group (27.3 words per minute). His word error rate across the task was 1.6%, and in 81% of sentence blocks, it was zero.

That result, from the BrainGate2 clinical trial at Massachusetts General Hospital, is the fastest brain-computer interface communication rate ever recorded using hand motor cortex decoding. It is 20 characters per minute faster than the previous record, set in 2021 using a handwriting-based approach. And it arrived the same month that China's National Medical Products Administration approved the world's first commercial invasive brain implant for patients with spinal cord injuries.

Two milestones, one conclusion: brain-computer interfaces are crossing from research curiosity to clinical product. But the companies raising billions are not the ones setting speed records. That disconnect is the real story.

The Speed Table Nobody Has Built

BCI communication speed has improved dramatically over the past decade, but the trajectory splits into two distinct tracks. Hand motor cortex decoding, which translates imagined hand or finger movements into text, has progressed from about 8 characters per minute to 110. Speech decoding, which reads the neural signals behind attempted speech, leapfrogged to 62 words per minute in a single study.

Year	System	Approach	Speed	Vocabulary	Error Rate
2017	BrainGate (Stanford)	Point-and-click cursor	~8 chars/min	Full QWERTY	Variable
2021	BrainGate (Stanford)	Handwriting decoding	90 chars/min (18 WPM)	Full alphabet	5.9% character
2023	UCSF (Willett/Chang)	Speech decoding	62 WPM	125,000 words	23.8% word (large vocab)
2024	Neuralink N1	Cursor control	Not disclosed (cursor, not text)	N/A	N/A
2026	BrainGate (MGH)	QWERTY touch-typing	110 chars/min (22 WPM)	Full keyboard + punctuation	0% word (81% of blocks)

Approaches diverge significantly. Speech decoding at UCSF hit 62 words per minute in 2023, nearly three times the new BrainGate typing record. But speech BCIs require the user to attempt vocalization, which is not possible for all patients. The ALS participant in the BrainGate study (T17) could not speak at all. He still typed at 47 characters per minute, faster than any previous system available to him. When researchers asked what he wanted to type freely, he recommended Zelda video games: "you should try oracle of ages and seasons... another is skyward sword... the music in those games is great." Three years after losing his voice to ALS, he was sharing game recommendations through his motor cortex.

How Fast Is Fast Enough?

Context makes the numbers concrete. Eye-tracking systems, the current standard of care for people with severe paralysis, produce roughly 5 to 10 words per minute and require sustained visual focus that fatigues users within 30 to 60 minutes. The BrainGate QWERTY system more than doubles that rate, and because it decodes imagined finger movements rather than eye position, participants can look away from the screen while typing.

Natural human conversation runs at about 160 words per minute. The UCSF speech decoder at 62 WPM is 39% of conversational speed. The BrainGate typing system at 22 WPM is 14%. Neither is fast enough for fluid conversation, but the typing system has a feature the speech decoder lacks: the user controls exactly when characters appear. No accidental leakage of private thoughts onto the screen. No AI interpretation of ambiguous neural signals generating words the user did not intend. As study author Daniel Rubin noted, participants have "full control of when, or when not, to share their thoughts."

One detail from the study deserves more attention: calibration took only 30 sentences. The decoder learned to translate T18's brain activity into accurate keystrokes after roughly 15 minutes of training. Previous BCI systems required hours or days of calibration per session. If this generalizes, it shrinks the setup burden from "requires a research team" to "requires a technician."

$3 Billion Chasing Six Companies

Six companies now control the commercial BCI industry, with a combined $2.4 billion in disclosed funding. Capital concentration is stark.

Company	Founded	Approach	Total Funding	Share	Clinical Status
Neuralink	2016	Intracortical array + surgical robot	$1.3B	54%	21 patients, 4 countries
Synchron	2012	Endovascular (Stentrode via jugular)	$345M	14%	Pivotal trial enrolled
Science Corp	2021	Retinal neuroprosthesis	~$270M	11%	Human trials (vision)
Blackrock Neurotech	2008	Utah Array (infrastructure)	$200M	8%	Research arrays in widespread use
Precision Neuroscience	2021	Subdural ECoG sheets	$155M	6%	Early human testing, IDE path
Paradromics	2015	High-throughput speech/intent	$118M+	5%	First human recording (2025)

Neuralink holds 54% of all disclosed BCI investment, and the top three companies control 80% of capital. That level of concentration in a pre-revenue industry is unusual. For comparison, the early autonomous vehicle sector spread investment across dozens of companies before consolidating. BCI started concentrated and is getting more so, with Synchron's $200 million Series D and Neuralink's $650 million Series E both closing in 2025.

Here is the part that should concern investors: speed records come from academic labs, not these companies. BrainGate, a research consortium spanning Massachusetts General Hospital, Brown University, Stanford, and Case Western Reserve, set the 22-WPM record. The 62-WPM speech record came from UCSF. Neuralink's public demonstrations have shown cursor control and gaming, not high-speed text communication. Synchron's endovascular approach avoids open brain surgery but records from the surface of blood vessels, which inherently captures fewer neurons than penetrating arrays.

China Got There First

On March 13, 2026, China's NMPA granted marketing approval to Neuracle Technology's NEO system, a brain-computer interface for adults with cervical spinal cord injuries. NEO integrates a cortical implant, an EEG electrode array, a neural signal transmitter, and a pneumatic robotic glove. Patients think about hand movements; the glove executes them. It is the first invasive BCI approved for commercial use anywhere in the world.

No equivalent device has received FDA clearance or CE marking. Neuralink is in clinical trials. Synchron is in pivotal trials. Precision is in early human testing. The U.S. pathway from first-in-human study to commercial approval for a Class III implanted neural device typically runs 5 to 8 years. China compressed that timeline, though transparency in the approval process and depth of safety data remain open questions. NMPA does not publish the equivalent of FDA's premarket approval summary documents.

Geopolitics makes this more than academic. BCIs are dual-use technology. A system that decodes intended hand movements from motor cortex could, with modifications, decode other cognitive states. China's military-civil fusion strategy explicitly targets brain-computer interface development as a priority area. The U.S. Defense Advanced Research Projects Agency (DARPA) has funded BCI research since the 1970s. The difference is that DARPA-funded work stays in labs. China's NEO is in hospitals.

The Strongest Case Against Excitement

Lab records do not survive contact with daily life. BrainGate participants typed with cables running from their skulls to external computers. The entire system requires a controlled clinical environment with trained staff. Neuralink's advantage is wireless transmission, which trades peak performance for usability. An implant that works at 22 WPM in a research suite but requires a percutaneous connector is not a product. An implant that works at 12 WPM wirelessly at home might be.

Signal degradation is the deeper concern. Utah Arrays used in BrainGate studies lose electrode recording quality over months to years as the brain's immune response encapsulates the foreign material. BrainGate reported data from sessions, not from continuous year-long use. Whether T18 can type at 22 WPM in year three with the same implant is unknown. Neuralink has reported thread retraction issues in its first patient. Synchron's endovascular approach may avoid the immune response problem but at the cost of signal resolution.

And then there is scale. Twenty-one Neuralink patients across four countries. Two BrainGate participants in one study. Synchron is in pivotal trials. The entire global population of humans with invasive BCIs is probably under 100. For context, approximately 1 million cochlear implants have been placed worldwide. BCIs are where cochlear implants were in the 1970s: working in labs, debated by ethicists, and years from routine clinical use.

What the Data Does Not Show

This analysis relies on published academic results and disclosed funding rounds. Several limitations apply. Speed records come from individual participants under controlled conditions and may not generalize. We do not have head-to-head comparisons across systems because each study uses different participants, conditions, and decoding algorithms. Neuralink's performance data is largely unpublished in peer-reviewed literature. Funding figures for private companies are self-reported and likely incomplete. China's NEO approval documents are not publicly available in English, making independent assessment of safety and efficacy data difficult. Long-term signal stability data beyond 2 to 3 years is sparse for all current-generation implants.

The Bottom Line

A man imagined typing on a keyboard, and a computer wrote what he intended at 81% of able-bodied speed. Not a demo. A peer-reviewed result from a registered clinical trial. BCI now has its proof-of-concept moment for high-speed, high-accuracy communication. What it lacks is a product. The companies with billions in funding are years behind the academic labs setting records, China has approved a commercial device the West cannot inspect, and every implant on the market will eventually be encapsulated by the body's immune system. The brain is readable. The question is whether any company can make that reading reliable, wireless, and lasting for a decade. Nobody has answered that yet.